We lifted 11 functions out of this paper's own repositories and ran 6 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| BlinkDL/RWKV-LM | canonical | 2 of 4 |
| blinkdl/chatrwkv | canonical | 1 of 3 |
| rwkv/rwkv.cpp | pwc_unofficial | 3 of 4 |
| Function | Status | Where it lives |
|---|---|---|
| batch | Ran | BlinkDL/RWKV-LM/RWKV-v7/train_temp/rwkv7_train_simplified.py code served (permissive licence) · get_code("7e9cf65e078afda5") |
| format_loss | Ran | rwkv/rwkv.cpp/python/measure_pexplexity.py code served (permissive licence) · get_code("922e0a890b2fbbab") |
| format_loss_with_perplexity | Ran | rwkv/rwkv.cpp/python/measure_pexplexity.py code served (permissive licence) · get_code("c64ea6a80f61aeac") |
| get_layer_count | Ran | rwkv/rwkv.cpp/python/convert_pytorch_to_ggml.py code served (permissive licence) · get_code("4dbb2fe483ceceed") |
| sample_logits | Ran | blinkdl/chatrwkv/RWKV_in_150_lines.py code served (permissive licence) · get_code("35e590faf9de67ca") |
| time_mixing__ | Ran | BlinkDL/RWKV-LM/RWKV-v7/rwkv_v7_demo_rnn.py code served (permissive licence) · get_code("07a576b3352ec5ac") |
| RUN_CUDA_RWKV7g | Not yet run | BlinkDL/RWKV-LM/RWKV-v7/train_temp/rwkv7_train_simplified.py code served (permissive licence) · get_code("4ba3321ca44d1608") |
| channel_mixing__ | Not yet run | BlinkDL/RWKV-LM/RWKV-v7/rwkv_v7_demo_rnn.py code served (permissive licence) · get_code("a43aabdbc73f2f29") |
| load_all_stat | Not yet run | blinkdl/chatrwkv/chat.py code served (permissive licence) · get_code("5256b8e5b92e6c98") |
| mm8 | Not yet run | blinkdl/chatrwkv/rwkv_pip_package/src/rwkv/model.py code served (permissive licence) · get_code("104b35c3b39791a5") |
| split_last_end_of_line | Not yet run | rwkv/rwkv.cpp/python/chat_with_bot.py code served (permissive licence) · get_code("ecd9b52290cb9d64") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Transformers have revolutionized almost all natural language processing (NLP) tasks but suffer from memory and computational complexity that scales quadratically with sequence length. In contrast, recurrent neural networks (RNNs) exhibit linear scaling in memory and computational requirements but struggle to match the same performance as Transformers due to limitations in parallelization and scalability. We propose a novel model architecture, Receptance Weighted Key Value (RWKV), that combines the efficient parallelizable training of transformers with the efficient inference of RNNs. Our approach leverages a linear attention mechanism and allows us to formulate the model as either a Transformer or an RNN, thus parallelizing computations during training and maintains constant computational and memory complexity during inference. We scale our models as large as 14 billion parameters, by far the largest dense RNN ever trained, and find RWKV performs on par with similarly sized Transformers, suggesting future work can leverage this architecture to create more efficient models. This work presents a significant step towards reconciling trade-offs between computational efficiency and model performance in sequence processing tasks.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2305.13048")
get_code_for_paper("2305.13048")
have("2305.13048")
Connect an agent — have() is free.